Anecdotes and educated guesses about why election officials make an unusual number of changes in polling places before an election are widespread. What's been lacking has been a procedure for measuring the impact of such changes and how different types of voters react. Fortunately, two political scientists have come up with a way to do this.
Writing in the current issue of the American Political Science Review (Feb 2011, Volume 105, No. 1), Henry Brady and John McNulty report their investigation of the 2003 gubernatorial recall election in California. (The APSR is behind a subscription wall, but you can find it at most college and university libraries). The election provided a natural experiment to investigate this question, as some counties attempted to cut costs by consolidating precincts and changing polling locations in ways that nearly randomly assigned higher voting costs to some but not all voters. By comparing turnout in the counties that did this with turnout in those that did not, the effects of the changes can be estimated.
Brady and McNulty began with voting records from the 2002 general election and the 2003 gubernatorial recall election in California. They had data on the residence of each voter, who voted, their polling place for each election, and their age. They had complete info on about 2.7 million voters.
They first analyzed whether there were any systematic predictors of which counties moved polling places. They found very weak systematic effects -- about 1% of the variation in polling place changes was due to factors such as the size of the precinct, the number of absentee voters, and the average age of voters. There was no significant effect of party of the registered voters in this case.
Their analysis of the voters found that precincts where polling places changed had an average decline of about 3% in polling place voting, an average increase of about 1.2% in absentee voting, and an average increase in non-voting of 1.9%. Of particular importance for Democrats, the average decline in Democrats' voting (2.1%), exceeded that for independents (1.8%) and Republicans (1.6%).
While there is no evidence of conscious manipulation in these California results, the findings suggest that it is appropriate to be watchful concerning the location of polling places. Areas experiencing notable population growth or decline are bound to experience some changes in polling places as a matter of routine, but wholesale changes, particularly when they appear to lack justification in administrative cost-saving, should be scrutinized quite closely.